LuxMark - The OpenCL CPU&GPU benchmark

[fellix_bg]

LuxMark is an OpenCL benchmark based on LuxRender. LuxRender is a physically based and unbiased rendering engine. Based on state of the art algorithms, LuxRender simulates the flow of light according to physical equations, thus producing realistic images of photographic quality - Gallery.

Info and Download

The default (start-up) benchmark mode is the GPU path.


Here is the my GTX570 (825/2100MHz):



Happy benchmarking!

[Toysoldier]

GTX580 (900/2004MHz)

[Toysoldier]

GTX580 (772/2004Mhz - default speed)

[Toysoldier]

GTX580 (825/2100Mhz)

[Sam_oslo]

Here is my undervalued (0.987v) 24/7 setup

GTX560 (910/2400MHz):

[NapalmV5]

900/2200 - 100mhz pcie

[kaktus1907]

@NapalmV5 sweet mother of mercy


OpenCL GPU Only = 5092
GTX 460 1GB @950/1900/4600MHz


Native CPU = 2393
Q9650 @4.00GHz

[zalbard]

2x 6970

[bamtan2]

very nice, very interesting. luxmark may have practical implications, unlike some synthetic benchmarks.

[freeloader]

And bringing up the bottom end just for fun.....AMD C50 1GZ dual core laptop.

[Roadhog]

Almost hit 20k on 3, 6970's @ 950/1450.

[STEvil]

GTX480 @ 800/2000 - 3800
E5450's @ 3.6ghz (8 cores) CPU Native - 3450

WinXP-64

Doesnt want to run on my Win7 system with the westmere's and 4870x2..

[Dade]

900/2200
30709 LuxMarks

Hi NapalmV5, I'm one of the LuxMark developer. Your score is the highest I have seen up to now. Cheers, you have redefined my idea of BFG9000

[STEvil]

hey dade any reason why the gtx480 gets half the score of a gtx570? does OS make that much difference, or could it be driver version?

[Dade]

hey dade any reason why the gtx480 gets half the score of a gtx570? does OS make that much difference, or could it be driver version?

Are you using the latest drivers ? NVIDIA has highly improved their OpenCL support in the latest release.

According Anandtech (http://www.anandtech.com/show/4135/n...-250-market/15), a 480gtx should score about the same value of a 570gtx with latest drivers.

If you check older Anandtech's reviews (http://www.anandtech.com/show/4061/a...eon-hd-6950/23), you can see like NVIDIA as nearly doubled the scores with latest drivers.

AMD has also improved the performance of HD6xxx family of a good 15-30% in the latest driver release.

[STEvil]

Thats exactly what it was I guess.

Went from 3800 with 260.69's to 8500 with 267.24's at 800/2000

[NapalmV5]

@NapalmV5 sweet mother of mercy

Hi NapalmV5, I'm one of the LuxMark developer. Your score is the highest I have seen up to now. Cheers, you have redefined my idea of BFG9000

thanks guys score goes to show how well youve implemented multi gpu support

950/2200 - 100mhz pcie

[jogshy]

After taking a look at the code, I wonder how dependent is this test on the PCI express. I've been never a big fan of passing tons of data by the bus.
Any of you can make a quick test pls? My BIOS does not allow to change the PCI multiplier... but I bet if you put a monster GPU and you put it to 2x a modest GPU running at 16x might be faster.

Also, I think the automatically assigned work size of 64 thread/block may cause an slowdown.

My ... strange.... results:

OpenCL - GPUs only ( a Radeon 5750, catalyst 11.2 ) -------------> 1784 marks


Native CPU ( an obsolete i7 920 ) ----> 3342 marks

So, in my case, the CPU is 2x faster than the GPU... Shouldn't be a GPU faster than a CPU rendering? That's the whole point of Octane, VRayRT and other GPGPU renderers after all...

[Brahmzy]

This is an awesome Benchmark - very useful.
I'll post some screens later today..

[Dade]

After taking a look at the code, I wonder how dependent is this test on the PCI express. I've been never a big fan of passing tons of data by the bus.
Any of you can make a quick test pls? My BIOS does not allow to change the PCI multiplier... but I bet if you put a monster GPU and you put it to 2x a modest GPU running at 16x might be faster.

Every 5 seconds, the images rendered by each GPU are downloaded to the CPU memory to be merged a drawn on the screen. It is a very little amount of data for the bandwidth of PCIe bus.
I have 2x5870 plus a 5850 installed on a 4x PCIe slot. I have done some test, moving the 5850 between 16x slots and 4x slot: I haven't experienced any performance difference.

Jogshy, anyway we need a feature like GPU-to-GPU direct transfer announced for CUDA 4.0 to solve this problem (if it becomes a problem with many GPUs).

Also, I think the automatically assigned work size of 64 thread/block may cause an slowdown.

A little trick: you can change that value by editing the file scenes/luxball/render-hdr.cfg. Just edit the following line:

opencl.gpu.workgroup.size = 64

So, in my case, the CPU is 2x faster than the GPU... Shouldn't be a GPU faster than a CPU rendering?

i7 920 is more expansive that a HD5750 and the GPU is still faster. I don't read it as a bad result for the CPUs Vs GPUs "holy war".

That's the whole point of Octane, VRayRT and other GPGPU renderers after all...

Jogshy, I assume you are the same of ompf.org forums; you know that there have been a lot of discussion about this point. Many people believe there is a lot of marketing vaporware in CPUs Vs GPUs "holy war".

My opinion is somewhat in the middle, GPUs computing offers great value (especially for home/office users) but it is no where near the 100x faster promised by the marketing.

[Toysoldier]

I tried changing the work group size to the suggested value of 640 and it throws an error
OpenCL ERROR: clEnqueueNDRangeKernel(-54)

So I tried with different work group sizes :

32 : OK ~ 7920
64 : OK ~ 8780
128: OK ~ 8650
256: OK ~ 8412
512: OK ~ 7580
640: OpenCL ERROR: clEnqueueNDRangeKernel(-54)
768: OpenCL ERROR: clEnqueueNDRangeKernel(-54)
1024: Rendering thread ERROR: clSetKernelArg(-38)

So it looks like 64 is the best workgroup size, at least on a GTX580 with the current codepath.

[jogshy]

i7 920 is more expansive that a HD5750 and the GPU is still faster. I don't read it as a bad result for the CPUs Vs GPUs "holy war".

Faster? My 5750 is actually 2x slower than the CPU as you can see in the screenshots.
Ok, the i920 is 2x more expensive than a 5750 so, relatively, the ATI GPU==CPU. A bit dissapointing.

Every 5 seconds, the images rendered by each GPU are downloaded to the CPU memory to be merged a drawn on the screen. It is a very little amount of data for the bandwidth of PCIe bus.

Taking a look to your code seems you're generating the rays in the CPU, then passing then to the GPU to perform the ray tests and then you bring back the results to the CPU. That implies a lot of PCI transfers, latency and synchronization. Can anybody with a decent BIOS test how much it scales, pls? My BIOS sux

[kaktus1907]

pcie bandwith numbers
16x pcie 2.0

Idle


Load

[jogshy]

pcie bandwith numbers

Nak, can your BIOS change the PCI multiplier?

[soya_crack]

Hi guys,

Do you know if a SLI Bridge is of advantage here?

@Napalm

Did you use one?

I am hitting 15k on 3x 560 GTX (not overclocked).